Keyboard Input Action Recognition Based On 3D Convolution

Computer vision technology,led by deep learning,has continued to progress,and has made great breakthroughs in image classification,video classification,target detection and other fields.Motion recognition refers to the technology of extracting and classifying motion features of video or image sequences through computer technology.Motion recognition has important significance in the fields of intelligent monitoring,video classification,and video content analysis.The paper attempts to integrate computer vision and information security,and explores the possibility of using visual technology to analyze the user's input on the keyboard from the video stream.For this task,the paper proposes a small data set.The method proposed in the paper first divides the video stream into small video segments and treats the video segments as three-dimensional images.In the paper,the target detection method is extended to the three-dimensional image,and the single-step model similar to the single-shot multi-frame detection is used to complete the recognition of the keyboard input action in the video clip.At the same time,the paper compares the basic single-shot multi-frame detection network,the single-shot multi-frame detection network using small convolution kernels,the single-shot multi-frame detection network using residual blocks,and the residual network using small convolution kernels in the paper task Performance.The experiment found that in the two evaluation indicators of the paper,the basic single-shot multi-frame detection network and the single-shot multi-frame detection network method using small convolution kernels have higher sequence accuracy,and the single-shot multi-frame detection using small convolution kernels The detection network has the highest average accuracy.Among them,the sequence accuracy directly related to the thesis task reaches a maximum of 89%.Therefore,the method of extracting the input sequence from the video stream through the three-dimensional convolutional neural network method is feasible.And it has a faster speed than human video analysis.Through proper optimization,real-time analysis can be done.This method can be applied to steal key information such as user keys.Therefore,in information security scenarios,it is necessary to take corresponding protective measures against such attacks from outside the system.